TalkChess.com

Hello Frank,

it works here with Ubuntu 8.10, gcc 4.3.2.

http://www.glaurungchess.com/glaurung22.zip We think our Plague plays better than Smaug. ;) Christopher I challange you and your silly co-threaders to take a program like Glaurung (but I can say also Toga) and increase it of 20 ELO points. You ARE not able to do it, otherwise you will be more respe...

Hi Gerd, my magic factor generation is much too experimental to give some code here. Most of the implementation details are just a guess what might work and you might better test your own ideas and come up with something better. Evaluation function (fitness function) First you need an evaluation fun...

Hi,

here 2 smaller magics for squares 52 and 53 found with a genetic optimization algorithm. The upper bits do NOT contain a shift count.
Greetings
Volker

Well, Volker, a trajectory is for me a possible beneficial combination. Reading your answer I see a problem: because your evaluation is depending on the prior move, and an identical situation nevertheless could be reached by other preceding moves, the reuse of result evaluations stored in the cache...

Interesting. Do you do anything special with the hashtable ( depth) ?

Hi Tony,

there is nothing special in the Hash table. I always store the full remaining depth in the hash table including the fractional part, but that has nothing to do with reducing the extensions away.

Greetings
Volker

What do you mean by promising trajectories? In Hermann I use fractional extensions. When a move is made, that is unrelated to the previous move and does not trigger an extension itself, the fractional part of the remaining depth is set to 0. A move is related to the previous one if - The moving piec...

I don't have a 64 bit compiler, so even if I installed Vista64 I wouldn't be able to check that aspect. The people here may be righ when they say that in 64 bit mode, it's not at all a problem. But, V C 2008 Express doesn't support 64 bit code, and I don't think spending $700 for the professional o...

Hello Gerd, Hermann no longer uses node counts for timing. (The public version 2.0 still does.) It depends too much on what one finds in the hash table and besides the draw problems you mentioned, it does not work very well with ponder on where after a ponder miss you may still have many hash hits i...

Hello Gerd, in Hermann I use neural networks for timing and for material evaluation. I have written something about timing in the old but still available newsticker at http://www.playwitharena.com . See Nr. 86 there. (You cannot put a direct link to the newsticker archive here.) The way I use neural...